Gitmore Launches 5 New Features: From AI Code Tracking to MCP Server

Developer reporting tool Gitmore has rolled out 5 new features at once, centered on two priorities: making code activity data easier for different roles to consume, and more proactively surfacing the new risks that come with AI's involvement in development.

The logic behind this update isn't complicated—when AI coding assistants start submitting PRs in bulk, "who changed what and when" becomes harder to track than before, and Gitmore is trying to provide a layer of aggregation and alerting in that gap.

Client-Facing Portals and Multilingual Reports

The first feature is Client Portals. Development teams can generate password-protected client portals to share project progress directly with non-technical audiences, without having to hand-write update emails each time.

Reports generate summaries based on audience type, support 8 languages, and can be exported as PDF. This design targets a long-standing friction point: engineering progress data is usually buried in Git commits and Issues, which clients or non-technical stakeholders find hard to read directly, often requiring someone to translate it in between. Gitmore automates that layer of translation.

Proactive Alerts: Watching AI-Written Code

The second feature, Proactive Alerts, is the most information-dense part of this update. It monitors three types of situations: stalled PRs, inactive AI agent PRs, and changes in the proportion of AI-written code.

Alerts are pushed via email or Slack. The third category is especially noteworthy—"changes in the ratio of AI-generated code" as a monitorable metric suggests teams are starting to treat AI involvement as an engineering variable that needs observation, rather than a pure efficiency tool. When this ratio suddenly rises or falls, it could mean something has gone wrong in the workflow, or it could mean code review pressure is building up.

"Ratio of AI-written code" as an engineering metric corresponds to the issue of code attribution. When an AI coding assistant directly produces code and submits a PR, Git's author field is often still recorded as the human developer who triggered it, so traditional author-based statistics can't distinguish between "human-written" and "AI-written" code. To obtain this ratio, you typically need to mark things at the commit message, PR metadata, or toolchain level, then aggregate by line count or commit count.

This metric is worth monitoring because it carries an assumed correlation with code review costs and defect rates: when the proportion of AI-generated code rises quickly in a short period, it often means PR throughput is up, the depth of each review is down, and latent technical debt and defects may be deferred in their exposure. Therefore, using changes in this ratio as an alert trigger is essentially approximating the hard-to-measure quantity of "review pressure" with a collectable proxy indicator.

Connecting to Notion, Linear, Jira, and Other Task Systems

The third feature is task management integration, supporting six platforms: Notion, Linear, Jira, Asana, ClickUp, and Trello.

There are two core capabilities: matching pull requests to the corresponding tasks, and automatically closing tasks after a PR is merged. Gitmore specifically emphasizes that any write-back operation requires human approval—clearly to avoid automated misoperations polluting the task system. For teams using multiple task tools at the same time, manually aligning PRs and tasks is a trivial but error-prone job, and automated matching can save a lot of time.

API and Read-Only MCP Server

The fourth feature is API and MCP (Model Context Protocol) support. Gitmore provides a read-only MCP server with 6 tools, allowing AI assistants like Claude to query repository activity data.

It also provides an API key for coding agents to call. MCP is a protocol standard promoted by Anthropic that lets AI assistants connect to external data sources in a unified way. Gitmore's choice of read-only mode is a prudent decision—letting AI query development insights but not giving it the ability to modify data, which makes the risk surface much smaller.

MCP (Model Context Protocol) is an open-source protocol standard released by Anthropic at the end of 2024, aimed at solving the problem of fragmented interfaces when AI assistants connect to external tools and data sources. Before a unified protocol existed, every time you wanted an AI assistant to read a new system (database, SaaS platform, code repository), developers had to write a separate integration logic for it, which was labor-intensive and hard to reuse. MCP's approach is to abstract the "data source" as an MCP Server and the "AI application" as an MCP Client, with both sides communicating in a unified message format. The Server exposes several tools externally, each with a clear name, parameter structure, and return format, and the model on the Client side decides which tool to call and what parameters to pass based on that.

MCP's key design is that permission boundaries are determined on the Server side. The Server can choose to implement only query-type tools without implementing write-type tools, so that even if the model makes a misjudgment, it cannot cause damage to external systems. The read-only MCP server provided by Gitmore (6 tools) is a typical implementation of this idea: letting assistants like Claude query repository activity data but not giving them any modification capability, narrowing the risk surface at the architectural level.

Report History Retained for One Year

The fifth feature is Report History. All sent reports are kept for one year, allowing you to look back at the original text of the email or Slack message sent at the time.

One detail is worth noting: viewing historical reports is not counted as an email open, because tracking pixels are removed before display. This small design is a plus for teams with compliance or privacy requirements.

Email open tracking pixel refers to an invisible 1×1 pixel image embedded in an HTML email, whose URL carries a unique identifier. When the recipient loads the email, the client requests this image from the server, and the server records that "the email has been opened" along with the time, number of opens, and approximate IP information. Marketing and customer success teams commonly use it to evaluate outreach effectiveness.

But tracking pixels also have side effects: they send information back without the recipient's knowledge, and some privacy regulations and corporate security policies block or strip such images. Gitmore removes tracking pixels before displaying historical reports, meaning that looking back at archives does not send signals back to anyone and is not counted as a new "open." For teams that need to retain communication records for audits, compliance, or customer dispute evidence, this ensures that archived content is semantically consistent with what was actually sent, rather than letting an internal review accidentally pollute open-rate data.

What This Update Says

Looking at these 5 features together, Gitmore's product direction points to a problem that is taking shape: after AI's participation in software development, the object of engineering observability has changed.

What used to need tracking was "what code people wrote," and now it also needs tracking "how much code AI wrote" and "whether anyone is managing the PRs AI submits." The monitoring of AI agent PRs in Proactive Alerts, as well as the tracking of AI code proportion, are both prepared for this new scenario. Client Portals and task integrations solve old problems—translating engineering data for non-engineering roles, and reducing manual shuttling between tools.

All features are live and available on the free plan. For teams that are bringing AI coding tools into their workflows, this set of Gitmore features offers an observation perspective worth noting.

One-line Summary

Gitmore launches 5 features, delivering both client-readable reports and monitoring of the new risks from AI's involvement in development.

Paragraph Summary

Gitmore has rolled out 5 new features at once, in two directions: making code activity data easier for different roles to consume, and more proactively surfacing the new risks from AI's involvement in development. Client Portals provides password-protected client portals, generates summaries by audience, supports 8 languages, and can export PDF; Proactive Alerts monitors stalled PRs, inactive AI agent PRs, and changes in the proportion of AI-written code, pushing alerts via email or Slack; task management integration connects Notion, Linear, Jira, Asana, ClickUp, and Trello, enabling PR-to-task matching and automatic closure after merge, with write-back operations requiring human approval; API and read-only MCP server let assistants like Claude query repository activity data; Report History retains reports for one year and removes tracking pixels when viewed to avoid counting as email opens. All features are live and available on the free plan.